पाठशाला Pathshala · उत्पाद Utpād, The product · Lesson 10 · Build
Pricing experiments that do not burn your customers
Price is the strongest lever most startups never test. Test it on new cohorts and separate pages, protect existing customers with grandfathering, and stay inside India’s rules on dark patterns and e-mandates.
Pathshala, The Founder Library · 11 October 2026 · 7 min read
In 2011 Netflix split its streaming and DVD service into two plans at $7.99 a month each, so a customer who wanted both now paid for two. In the quarter that followed its US subscriber count fell by 800,000, to about 23.8 million. The price may well have been right. The way it reached existing customers was not, and that is the part of pricing a startup can control.
Price is usually the strongest lever on revenue a young company has and the one it tests least, because founders fear exactly that kind of revolt. This lesson gives a way to learn what customers will pay without putting existing customers through an experiment they did not agree to: what a price test may touch, three instruments that are safe, the arithmetic of a rise, how big a test must be to mean anything, the Indian rules a price change has to respect, and how to move existing customers when the time comes.
What a price test is allowed to touch
A price an existing customer agreed to is a promise, and a test that breaks it is not an experiment; it is a price rise without the courtesy. The working rule is therefore simple. Test on people who have not yet bought. Prospects can be shown different offers because they have no agreement to break, as long as every offer is real and the price shown is the price charged. Existing customers change price only by a decision, announced in advance, explained and usually phased.
Two more rules keep tests honest. Never show two people in the same conversation different prices for the same thing; in India, where a team’s customers talk to one another in trade associations and WhatsApp groups, that conversation happens faster than founders expect. And never test by hiding cost: a lower headline with fees added at checkout is not a price test, it is a dark pattern, and since 2023 it has had a name in Indian rules.
Three instruments that do not burn anyone
Cohort tests. Change the price for everyone who signs up after a date, leave everyone before it alone, and compare the two cohorts on trial starts, conversion to paid, revenue per sign-up and month-three retention. It is slower than a split test because the cohorts differ in season and mix, but nobody sees two prices at once, and it is what most companies do in practice. Run each price for at least a full sales cycle and compare like with like: the same channels, the same month of the year where you can.
Separate pages for separate segments. Build distinct landing pages for distinct audiences, each with its own offer: clinics in tier-two cities and hospital chains, a self-serve plan for small firms and an annual plan for larger ones, a city where you are launching and one where you are established. Each page is consistent for everyone who sees it, and the comparison across pages tells you about willingness to pay by segment, which is usually what you needed to know. Painted doors belong here too: a higher tier listed with a button that leads to a short form saying it is coming, to measure how many click before you build it.
Grandfathering. When a new price is adopted, existing customers keep theirs for a stated period, often twelve months, and are moved later with notice. This is what lets the first two instruments run at all, because it removes the fear that a test on new customers will leak to old ones. Its cost is a slower rise in revenue, which is cheap next to the cost of a revolt.
The arithmetic of a rise
When the time comes to move existing customers, one line of arithmetic tells you how much loss a rise can absorb. If the price rises by r and a share c of customers leave, revenue becomes (1 + r) × (1 − c) of what it was. That is above today as long as c is below r ÷ (1 + r). A 10 per cent rise can lose 9.1 per cent of customers and break even; a 20 per cent rise, 16.7 per cent; a 50 per cent rise, a third.
The break-even looks generous, and that is the trap. It counts only this month’s revenue. A customer who leaves takes every future month with them, the customers who stay at a higher price may leave sooner, and an angry customer in a trade group costs more than one subscription. Use the figure the other way round: decide the largest loss you would accept, read off the rise it allows, then set the guess at who leaves at the bad case rather than the hopeful one.
How many customers a test needs
Price tests on small traffic usually prove nothing, and it is better to know before starting. The standard approximation for comparing two conversion rates at 5 per cent significance and 80 per cent power is sixteen times p × (1 − p), divided by the square of the difference you want to detect, for each arm. With 3 per cent of visitors becoming paying customers and a wish to detect a fall to 2.4 per cent, that is about 13,000 visitors for each price. Evan Miller’s calculator gives exact figures for your own numbers.
A company with 2,000 visitors a week cannot run that test in a sensible time. It has three better options. Test larger differences, ₹999 against ₹1,499 rather than ₹999 against ₹1,099, because a big gap needs far fewer visitors to show. Measure revenue per visitor, which is what you actually care about and which a higher price can win even while conversion falls. Ask before you test: the [customer interview](/library/the-customer-interview-done-properly) and a written quote to twenty prospects at the higher price will tell you more at low volume than a split test that never reaches significance.
Run the experiment on people who have not bought yet. Make the decision for people who have, and tell them before it happens.
The Indian rulebook for a price change
Two sets of rules apply to most consumer and small-business software sold in India; both were checked in October 2026 and should be checked again before a change. The first is the Central Consumer Protection Authority’s Guidelines for Prevention and Regulation of Dark Patterns, issued on 30 November 2023. They specify thirteen patterns, among them false urgency, drip pricing, bait and switch, subscription trap and SaaS billing. In June 2025 the authority asked e-commerce platforms to audit themselves for these patterns within three months. A price test that shows one price on the page and adds to it at checkout, or advertises an offer that is not available when the customer arrives, is no longer a growth tactic. Every arm of a test must be honest on its own.
The second is RBI’s framework for recurring payments by e-mandate. Its 2019 circular on cards requires additional authentication when a mandate is registered, a notice to the customer at least 24 hours before each debit with the amount, and a maximum amount set by the customer for variable charges; any change to an existing mandate also needs additional authentication. Recurring debits without that authentication are capped per transaction; the cap was ₹15,000 for most categories when a December 2023 circular raised it to ₹1 lakh for mutual funds, insurance premiums and credit card bills. The practical consequence: a price rise that takes a charge above the customer’s mandate maximum needs the customer to approve again, and every re-approval is a moment at which some customers do not. Plan the rise around it, with notice and a link to re-authorise, rather than discovering it as a wave of failed payments.
Grandfathering and the migration, done properly
When existing customers do move, five things keep it from becoming a revolt. Notice: at least thirty days, and sixty for annual or high-value accounts. A reason that is true and specific: what has been added since they signed up, what it costs to serve them. A bridge: a period at the old price, or a discount for paying annually now, so the customer has a choice rather than an ultimatum. A conversation with the twenty largest accounts before the email, by phone, from the founder. A record of every customer who complains and what they were offered, so nobody receives a better deal for being louder without a decision to give it. Then hold the new price. A rise withdrawn after a week of complaints teaches customers that complaining works.
A pricing test, week by week
Week one: write the hypothesis, the price, the segment, the metric (revenue per visitor or per sign-up) and the smallest result that would change your mind; check the sample against your traffic and widen the gap if it is too small. Week two: build the page or set the cohort date, and check every arm shows the price it charges with nothing added at checkout. Weeks three to eight: run without peeking at daily results, and talk to every prospect who drops at the pricing page. Week nine: read the result and decide. Each quarter, run one price test on new customers. Each year, review existing customers’ prices and decide, with notice, who moves and when. Keep a log of every test, its result and what you did; it is the pricing strategy the company actually has.
Nothing here is legal, tax or investment advice. The rules on dark patterns and e-mandates change; read the current text before a change goes live.
Sources
- TV Technology, Netflix reports loss of 800,000 domestic subscribers in Q3, October 2011
- PIB, CCPA issues Guidelines for Prevention and Regulation of Dark Patterns, 2023, listing 13 specified dark patterns, December 2023
- PIB, CCPA advisory to e-commerce platforms to self-audit for dark patterns within 3 months, June 2025 — Lists the thirteen patterns by name. Checked October 2026.
- RBI, Processing of e-mandate on cards for recurring transactions, circular of 21 August 2019
- RBI, Processing of e-mandates for recurring transactions, circular of 12 December 2023 — Raises the limit without additional authentication from ₹15,000 to ₹1 lakh for three categories. Checked October 2026.
- Evan Miller, Sample Size Calculator